src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_pd(fr->ic->k_rf);
 107     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_pd(fr->ic->c_rf);
 109
 110     /* Avoid stupid compiler warnings */
 111     jnrA = jnrB = 0;
 112     j_coord_offsetA = 0;
 113     j_coord_offsetB = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 117
 118     /* Start outer loop over neighborlists */
 119     for(iidx=0; iidx<nri; iidx++)
 120     {
 121         /* Load shift vector for this list */
 122         i_shift_offset   = DIM*shiftidx[iidx];
 123
 124         /* Load limits for loop over neighbors */
 125         j_index_start    = jindex[iidx];
 126         j_index_end      = jindex[iidx+1];
 127
 128         /* Get outer coordinate index */
 129         inr              = iinr[iidx];
 130         i_coord_offset   = DIM*inr;
 131
 132         /* Load i particle coords and add shift vector */
 133         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 134
 135         fix0             = _mm_setzero_pd();
 136         fiy0             = _mm_setzero_pd();
 137         fiz0             = _mm_setzero_pd();
 138
 139         /* Load parameters for i particles */
 140         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 141
 142         /* Reset potential sums */
 143         velecsum         = _mm_setzero_pd();
 144
 145         /* Start inner kernel loop */
 146         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 147         {
 148
 149             /* Get j neighbor index, and coordinate index */
 150             jnrA             = jjnr[jidx];
 151             jnrB             = jjnr[jidx+1];
 152             j_coord_offsetA  = DIM*jnrA;
 153             j_coord_offsetB  = DIM*jnrB;
 154
 155             /* load j atom coordinates */
 156             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 157                                               &jx0,&jy0,&jz0);
 158
 159             /* Calculate displacement vector */
 160             dx00             = _mm_sub_pd(ix0,jx0);
 161             dy00             = _mm_sub_pd(iy0,jy0);
 162             dz00             = _mm_sub_pd(iz0,jz0);
 163
 164             /* Calculate squared distance and things based on it */
 165             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 166
 167             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 168
 169             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 170
 171             /* Load parameters for j particles */
 172             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 173
 174             /**************************
 175              * CALCULATE INTERACTIONS *
 176              **************************/
 177
 178             /* Compute parameters for interactions between i and j atoms */
 179             qq00             = _mm_mul_pd(iq0,jq0);
 180
 181             /* REACTION-FIELD ELECTROSTATICS */
 182             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 183             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 184
 185             /* Update potential sum for this i atom from the interaction with this j atom. */
 186             velecsum         = _mm_add_pd(velecsum,velec);
 187
 188             fscal            = felec;
 189
 190             /* Update vectorial force */
 191             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 192             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 193             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 194
 195             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 196                                                    _mm_mul_pd(dx00,fscal),
 197                                                    _mm_mul_pd(dy00,fscal),
 198                                                    _mm_mul_pd(dz00,fscal));
 199
 200             /* Inner loop uses 35 flops */
 201         }
 202
 203         if(jidx<j_index_end)
 204         {
 205
 206             jnrA             = jjnr[jidx];
 207             j_coord_offsetA  = DIM*jnrA;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_pd(ix0,jx0);
 215             dy00             = _mm_sub_pd(iy0,jy0);
 216             dz00             = _mm_sub_pd(iz0,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 220
 221             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 222
 223             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 224
 225             /* Load parameters for j particles */
 226             jq0              = _mm_load_sd(charge+jnrA+0);
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm_mul_pd(iq0,jq0);
 234
 235             /* REACTION-FIELD ELECTROSTATICS */
 236             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 237             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 238
 239             /* Update potential sum for this i atom from the interaction with this j atom. */
 240             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 241             velecsum         = _mm_add_pd(velecsum,velec);
 242
 243             fscal            = felec;
 244
 245             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 246
 247             /* Update vectorial force */
 248             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 249             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 250             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 251
 252             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 253                                                    _mm_mul_pd(dx00,fscal),
 254                                                    _mm_mul_pd(dy00,fscal),
 255                                                    _mm_mul_pd(dz00,fscal));
 256
 257             /* Inner loop uses 35 flops */
 258         }
 259
 260         /* End of innermost loop */
 261
 262         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 263                                               f+i_coord_offset,fshift+i_shift_offset);
 264
 265         ggid                        = gid[iidx];
 266         /* Update potential energies */
 267         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 268
 269         /* Increment number of inner iterations */
 270         inneriter                  += j_index_end - j_index_start;
 271
 272         /* Outer loop uses 8 flops */
 273     }
 274
 275     /* Increment number of outer iterations */
 276     outeriter        += nri;
 277
 278     /* Update outer/inner flops */
 279
 280     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*35);
 281 }
 282 /*
 283  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_double
 284  * Electrostatics interaction: ReactionField
 285  * VdW interaction:            None
 286  * Geometry:                   Particle-Particle
 287  * Calculate force/pot:        Force
 288  */
 289 void
 290 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_double
 291                     (t_nblist                    * gmx_restrict       nlist,
 292                      rvec                        * gmx_restrict          xx,
 293                      rvec                        * gmx_restrict          ff,
 294                      t_forcerec                  * gmx_restrict          fr,
 295                      t_mdatoms                   * gmx_restrict     mdatoms,
 296                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 297                      t_nrnb                      * gmx_restrict        nrnb)
 298 {
 299     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 300      * just 0 for non-waters.
 301      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 302      * jnr indices corresponding to data put in the four positions in the SIMD register.
 303      */
 304     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 305     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 306     int              jnrA,jnrB;
 307     int              j_coord_offsetA,j_coord_offsetB;
 308     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 309     real             rcutoff_scalar;
 310     real             *shiftvec,*fshift,*x,*f;
 311     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 312     int              vdwioffset0;
 313     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 314     int              vdwjidx0A,vdwjidx0B;
 315     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 316     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 317     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 318     real             *charge;
 319     __m128d          dummy_mask,cutoff_mask;
 320     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 321     __m128d          one     = _mm_set1_pd(1.0);
 322     __m128d          two     = _mm_set1_pd(2.0);
 323     x                = xx[0];
 324     f                = ff[0];
 325
 326     nri              = nlist->nri;
 327     iinr             = nlist->iinr;
 328     jindex           = nlist->jindex;
 329     jjnr             = nlist->jjnr;
 330     shiftidx         = nlist->shift;
 331     gid              = nlist->gid;
 332     shiftvec         = fr->shift_vec[0];
 333     fshift           = fr->fshift[0];
 334     facel            = _mm_set1_pd(fr->epsfac);
 335     charge           = mdatoms->chargeA;
 336     krf              = _mm_set1_pd(fr->ic->k_rf);
 337     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 338     crf              = _mm_set1_pd(fr->ic->c_rf);
 339
 340     /* Avoid stupid compiler warnings */
 341     jnrA = jnrB = 0;
 342     j_coord_offsetA = 0;
 343     j_coord_offsetB = 0;
 344
 345     outeriter        = 0;
 346     inneriter        = 0;
 347
 348     /* Start outer loop over neighborlists */
 349     for(iidx=0; iidx<nri; iidx++)
 350     {
 351         /* Load shift vector for this list */
 352         i_shift_offset   = DIM*shiftidx[iidx];
 353
 354         /* Load limits for loop over neighbors */
 355         j_index_start    = jindex[iidx];
 356         j_index_end      = jindex[iidx+1];
 357
 358         /* Get outer coordinate index */
 359         inr              = iinr[iidx];
 360         i_coord_offset   = DIM*inr;
 361
 362         /* Load i particle coords and add shift vector */
 363         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 364
 365         fix0             = _mm_setzero_pd();
 366         fiy0             = _mm_setzero_pd();
 367         fiz0             = _mm_setzero_pd();
 368
 369         /* Load parameters for i particles */
 370         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 371
 372         /* Start inner kernel loop */
 373         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrA             = jjnr[jidx];
 378             jnrB             = jjnr[jidx+1];
 379             j_coord_offsetA  = DIM*jnrA;
 380             j_coord_offsetB  = DIM*jnrB;
 381
 382             /* load j atom coordinates */
 383             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 384                                               &jx0,&jy0,&jz0);
 385
 386             /* Calculate displacement vector */
 387             dx00             = _mm_sub_pd(ix0,jx0);
 388             dy00             = _mm_sub_pd(iy0,jy0);
 389             dz00             = _mm_sub_pd(iz0,jz0);
 390
 391             /* Calculate squared distance and things based on it */
 392             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 393
 394             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 395
 396             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 397
 398             /* Load parameters for j particles */
 399             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq00             = _mm_mul_pd(iq0,jq0);
 407
 408             /* REACTION-FIELD ELECTROSTATICS */
 409             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 410
 411             fscal            = felec;
 412
 413             /* Update vectorial force */
 414             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 415             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 416             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 417
 418             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 419                                                    _mm_mul_pd(dx00,fscal),
 420                                                    _mm_mul_pd(dy00,fscal),
 421                                                    _mm_mul_pd(dz00,fscal));
 422
 423             /* Inner loop uses 30 flops */
 424         }
 425
 426         if(jidx<j_index_end)
 427         {
 428
 429             jnrA             = jjnr[jidx];
 430             j_coord_offsetA  = DIM*jnrA;
 431
 432             /* load j atom coordinates */
 433             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 434                                               &jx0,&jy0,&jz0);
 435
 436             /* Calculate displacement vector */
 437             dx00             = _mm_sub_pd(ix0,jx0);
 438             dy00             = _mm_sub_pd(iy0,jy0);
 439             dz00             = _mm_sub_pd(iz0,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 443
 444             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 445
 446             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 447
 448             /* Load parameters for j particles */
 449             jq0              = _mm_load_sd(charge+jnrA+0);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq00             = _mm_mul_pd(iq0,jq0);
 457
 458             /* REACTION-FIELD ELECTROSTATICS */
 459             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 464
 465             /* Update vectorial force */
 466             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 467             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 468             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 469
 470             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 471                                                    _mm_mul_pd(dx00,fscal),
 472                                                    _mm_mul_pd(dy00,fscal),
 473                                                    _mm_mul_pd(dz00,fscal));
 474
 475             /* Inner loop uses 30 flops */
 476         }
 477
 478         /* End of innermost loop */
 479
 480         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 481                                               f+i_coord_offset,fshift+i_shift_offset);
 482
 483         /* Increment number of inner iterations */
 484         inneriter                  += j_index_end - j_index_start;
 485
 486         /* Outer loop uses 7 flops */
 487     }
 488
 489     /* Increment number of outer iterations */
 490     outeriter        += nri;
 491
 492     /* Update outer/inner flops */
 493
 494     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 495 }